The 2026 memory price surge, in the makers' own numbers

The same shape appears at each of the makers below, and the scale differs by how much of each business is memory.

Two arithmetic notes, both ours rather than the companies’. Samsung’s DS division operating profit of KRW 89.2 trillion is 99.7% of the group’s KRW 89.5 trillion — on those published figures, essentially the entire company’s operating profit came from the division that sells memory and chips 3. And Winbond’s gross margin rose 12.8 percentage points between its own two published figures, 53.4% in Q1 and 66.2% in Q2 7 6.

One oddity recorded rather than explained, because no source read for this page explains it. SK hynix reports "93.9226 trillion won in net profit (with a net margin of 118%)" for a quarter in which operating profit was 60.5426 trillion won 4 — that is, net profit above operating profit, and a net margin above 100%. The announcement states both figures and gives no reconciliation of the gap. This page reports what the disclosure says and does not speculate about what sits between the two lines.

The cleanest confirmation comes from a smaller company, because its numbers are simpler. Nanya Technology states that its bit shipment was flat while its price rose 5:

A company shipping the same number of bits as three months earlier, for more than sixty percent more money, is as close to an experimental control as this subject offers. Whatever else was happening in 2026, it was not that the makers found a way to produce dramatically more memory.

SK hynix says the same thing in words rather than figures: "Both DRAM and NAND flash memory prices experienced significant quarter-over-quarter increases", in a quarter where, it says, "high-performance products for AI servers led price increases" 4. Samsung likewise attributes its record to "higher average selling price (ASP)" and to "industry-wide memory price increases" 8.

One figure that needs its qualifier kept attached, because dropping it changes the claim. Winbond states: "Blended ASP increased by approximately 100% QoQ, reflecting both higher pure ASP and a favorable product mix." 6 The second half of that sentence is Winbond’s own, and it matters — a blended average selling price rises when the mix shifts toward more expensive products, which is not the same as every product costing twice as much. Winbond says in the same release that its Customized Memory Solution line "accounted for 53% of Q2 2026 revenue, primarily driven by strong AI-related demand" 6. Anyone quoting the 100% without the mix clause is quoting half a sentence.

You can date the turn from Samsung’s own language, quarter by quarter, because it changes.

The wording moves from "favorable" to "surge" to "continued…upward trend" across four quarters, and Samsung’s group operating profit moves with it: KRW 12.2 trillion in Q3 2025, KRW 20.1 trillion in Q4 2025, KRW 57.2 trillion in Q1 2026, KRW 89.5 trillion in Q2 2026 10 11 8 3.

The turn was also anticipated in writing before it fully arrived, which is unusual and worth recording.

SK hynix published a market outlook on 5 January 2026 built around what it called a supercycle, noting that "The term “supercycle” has been used in the industry to describe the strong momentum in the memory sector since 2024" 9. In the same piece it relays a forecast it attributes to a trade-statistics body rather than making it itself: "According to the World Semiconductor Trade Statistics (WSTS), the global semiconductor market will grow by more than 25% year-over-year in 2026, reaching approximately $975 billion, with the memory segment increasing at 30% growth." 9 That attribution clause is part of the sentence and is kept here, because a forecast SK hynix passes on is a different thing from a forecast SK hynix makes.

Two months earlier, a much smaller company had been more specific about which parts would tighten. Etron Technology’s November 2025 investor conference described a "shortage in memory supply, a situation that is unlikely to ease in the near term", and named the products: "specific products such as DDR4, DDR3, and LPDDR4/4X are expected to remain in tight supply until the second half of next year, or even continue to 2027" 12. Those are the older memory types — the ones in machines people already own and want to upgrade.

By January 2026 the foundry side was moving too. PSMC announced it had "followed the signing of a Letter of Intent (LOI) with Micron Technology by immediately launching a DRAM process advancement plan", describing a "global DRAM market’s long-term supply deficit driven by the explosion of AI applications" and saying it was "capitalizing on opportunities presented by supply shortages and rising prices" 13.

A causal claim this page states carefully, because the usual version of it overshoots. It is widely written that the makers discontinued DDR4 in this period, and that this is why older memory rose fastest. No maker’s own announcement of a 2025 or 2026 DDR4 discontinuation or output reduction was found for this page. What does exist is older and narrower: Micron issued product change notice 34165 on 2 June 2021, titled "End of Life (EOL) of Select DDR3, DDR4 and RLDRAM Products", stating that "Micron will discontinue select DDR3, DDR4 and RLDRAM components and modules." 14 That is a 2021 notice covering select parts, four years before the surge, and it will not carry the weight the popular version puts on it. Capacity being diverted toward AI products is documented in the makers’ own words above; a wholesale DDR4 cancellation during the surge is not, and the two should not be run together.

Within the makers, the spread comes down to how much of the business is the thing whose price moved.

Nanya is nearly pure DRAM, and its numbers are the most violent: ASP up more than 500% year on year, an operating margin of 73.7%, on net sales of NT$82,549 million 5. Micron sells DRAM and NAND, both of which surged, and its filed figures work out to a gross margin of 84.56% — our arithmetic; Micron prints 84.6% 1 2. Samsung is a conglomerate, so the group figure understates what happened inside it — which is why the DS division line matters, at KRW 127.5 trillion of the group’s KRW 171.5 trillion in revenue 3.

Winbond is the instructive middle case, because it publishes the mix that produced its number. Its Q2 2026 revenue of NT$59.843 billion was 53% Customized Memory Solution, "primarily driven by strong AI-related demand", and its blended ASP rise of "approximately 100% QoQ" is explicitly attributed to "both higher pure ASP and a favorable product mix" 6. A company whose AI-facing line accounted for over half its revenue that quarter is not a clean read on what happened to commodity memory prices.

Two things worth noticing in the shape of the data rather than its level. Etron’s own month-on-month figures run 22.39, 13.81, 26.85, 20.91, 24.75, 14.40, 11.70 and 6.36 from January to August 16 — not a steady deceleration, since March and May are both above January, but a clear one from May onward, which is what the late stage of a price move can look like from inside. And the whole picture is a snapshot: Nanya had not published an August 2026 monthly figure when this page was checked 17, and Micron’s fiscal fourth quarter was not yet reported. Nothing here should be read as the end of the story, because the story was still being filed.

What the surge did to demand, rather than to price, shows up in Corsair’s numbers. Over the three months to 30 June 2026, its memory products revenue was $123,014 thousand against $104,973 thousand a year earlier, while total net revenue fell from $320,112 thousand to $314,335 thousand 19.

So memory revenue rose about 17.2% while the company as a whole shrank about 1.8% — both our arithmetic on Corsair’s published figures. Corsair itself states the memory figure as "memory revenue grew 17% year-over-year" 20, which matches.

Its chief executive, Thi La, put the demand effect on the record: "While elevated memory pricing continues to delay DIY PC builds, we believe that demand has been deferred, not lost, and that Corsair is well positioned to benefit as market conditions normalize." 20 The same release says "Higher memory pricing continued to weigh on the market for DIY PC builds" 20.

Why a module brand’s accounts are a good thermometer here. Corsair’s quarterly filing explains the structural fact that makes it one: under the heading "Impact of Fluctuations in Integrated Circuits Pricing", it states that integrated circuits "account for most of the cost of producing our high-performance memory products" 19, and its annual filing’s risk factors note that DRAM ICs "account for most of the cost of producing our DRAM modules" and that their prices "have historically been subject to volatility over relatively short periods of time" 21. A module brand buys the expensive part and adds a board. When the chip price triples, almost the whole of that lands on the module brand’s cost line — which is exactly why the eight module-brand guides in this section matter more in a year like this one.

One further effect is visible in Micron’s filings and is worth understanding, because it changes who gets supply in a shortage.

Micron has been signing what it calls strategic customer agreements. From its quarterly filing: "These agreements are structured as take-or-pay agreements, with binding commitments for specific volumes over the multi-year contract terms" and "Pricing for most agreements is either fixed, or is subject to minimum and maximum pricing." 1 A subheading on its earnings release calls them "transformational" 2, and its chief executive said: "We believe our multi-year Strategic Customer Agreements will significantly enhance the durability and predictability of Micron’s strong financial performance." 2 That is a stated belief about the future by an interested party, and is recorded here as one.

What a take-or-pay structure does, by its nature, is commit volume in advance. A customer commits to buy specified volumes over years; the maker commits to supply them. That is the ordinary meaning of the term and of the clause Micron quotes, though Micron does not spell out the consequence for anyone outside such an agreement. Micron states that it expects "to receive cash deposits and related financial commitments of $22 billion for agreements concluded to date" 1. Note the wording: deposits and commitments, which are not the same thing, and the filing sentence quoted here does not say how the $22 billion divides between them.

Stated carefully, because the inference is easy to overdraw. No source read for this page says that these agreements reduce supply to India, to retail, or to any particular market. What the documents establish is the structure: multi-year binding volume commitments, largely pre-paid, held by large customers. What the documents establish is the structure: multi-year binding volume commitments held by large customers, alongside a separate and explicit decision to leave the consumer channel "to improve supply and support for our larger, strategic customers in faster-growing segments" 18. Those are two documented facts, not a proven chain between them, and the size of any effect on an individual buyer is neither published nor estimated here.

Two counterweights belong here, because a page that only stacks up bullish quotations is not describing a cycle.

The first is that the makers themselves say this reverses. Micron’s own risk factors carry the heading "Volatility in average selling prices for our semiconductor memory and storage products may adversely affect our business.", and state: "Increases in worldwide supply of semiconductor memory and storage, if not accompanied by commensurate increases in demand, could lead to declines in average selling prices for our products and could materially adversely affect our business, results of operations, or financial condition." 23 The companies above are spending heavily to do precisely that — increase worldwide supply. The mechanism that ends a memory shortage is the same one that causes the next glut, and Micron keeps that warning in its filings through the whole cycle: it appears in the annual report for the year ended 28 August 2025 23, a year whose own gross margin worked out to 39.79% on the figures in that filing, and the quarterly filings that followed carry the same risk in their summaries 1.

The second is that the industry's own literature flags a correction. SK hynix’s market outlook notes, without endorsing it: "Some market researchers and overseas media suggest that HBM prices could enter a correction phase after 2026 due to intensified competition and expanded production capacity." 9 That is a second-hand view reported by SK hynix, not SK hynix’s own forecast, and this page presents it as exactly that.

How to read the guidance honestly. Corsair’s chief executive frames the demand side as "deferred, not lost", expecting benefit "as market conditions normalize" 20 — a company whose business improves when memory gets cheaper, saying it expects memory to get cheaper. Samsung, whose business improves when memory stays expensive, projects the market to stay "undersupplied" 3. Both are forecasts by interested parties and neither is evidence. The construction schedules are not forecasts, which is why they carry the weight in this section.

Three further limits that affect how the figures on this page can be compared with each other.

The fiscal calendars do not line up. Micron’s "third quarter of 2026" ended 28 May 2026 1; Samsung’s and SK hynix’s second quarters ended in June; the Taiwanese companies report on calendar months. A table that puts them in one row, as this page does, is aligning approximately and should be read that way.

The currencies are not converted anywhere on this page. Samsung and SK hynix are quoted in won; Nanya, Winbond, PSMC and Etron in New Taiwan dollars, except where a company itself uses another — Nanya states its long-run capex plan in US dollars 5; Micron and Corsair in US dollars. Converting them would mean choosing an exchange rate and a date, and would turn published figures into derived ones. Each figure is left in the currency its source uses.

Some disclosures were not readable and some did not exist yet. Micron’s fiscal fourth quarter had not been reported when this page was checked; Nanya had not published an August 2026 monthly figure 17; and several long filings could only be read in part. Where a figure here is our arithmetic rather than a company’s, the page says so at the point of use.

The honest summary of what this page is. It is not a price guide and cannot be used as one. It is an account of a price event assembled entirely from what the companies involved told regulators and shareholders about themselves — which is a narrower thing than the coverage elsewhere, and the only kind this site is willing to publish. Where the filings are silent, this page is silent too, and says which silence it is.